From Data to Discovery: Machine Learning in High Energy Physics at CERN Experiments

Vortrag von Dr. Shah Rukh Qasim

Datum: 21.11.24  Zeit: 12.15 - 13.45  Raum: Y27H12

At the Large Hadron Collider (LHC) at CERN, high energy protons collide at the center of various experiments at a rate of 40 million multi-particle collisions per second, amounting to petabytes of data annually. A complex infrastructure is required to select and store some of this data, and machine learning is used to choose the events that might be interesting to look at for offline storage. In each of these events, thousands of secondary particles are produced, which leave their 3D image in complex experiments (such as the CMS and the ATLAS experiment) composed of various detectors. Reconstructing back the particles from the 3D pictures is a monumental task due to the sheer scale as well as the irregularity of various detector elements, necessitating the use of advanced graph neural networks. This task is similar to object detection or semantic segmentation in computer vision. After reconstructing these events, the data is compared to that from standard theories, obtained via Monte Carlo simulations. However, these precise simulators are very computationally expensive. Generative machine learning, known for its power to be able to generate realistic images and videos, is now also being used as a faster surrogate to complex simulations. Finally, machine learning is also being used to design complex physics instruments. As an example, the design of the Muon Shield of the new SHiP experiment is being optimized with machine learning. This talk will provide a comprehensive overview of using machine learning for the above-mentioned applications, along with the challenges involved and the specific methodologies required.